An Empirical Analysis Of P2P Loan Default Prediction Models

With the growing of Internet technology and the innovation of finance,petty loan is combined with Internet,which is called P2 P lending.P2 P lending is a good complement of retail loan in bank.A certain borrower who has financial requirement and cannot borrow money from bank can borrow money via P2 P platform.Because P2 P lending is more convenient than retail loan,the quantity of P2 P platforms and the trading volume grows rapidly.However,there are some problems in P2 P industry recently and a lot of P2 P companies go bankrupt or run off with money.It cannot be denied that some of them have problems like illegal funding,Ponzi scheme and so on,but most of them is caused by high default rate and weak risk management.Maintaining low default rate is very important for P2 P companies and is the foundation of the development of P2 P companies.Financial industry puts great emphasis on risk management and P2 P industry is no exception.Other than traditional retail loans,the customers of P2 P loan cannot meet the requirements of banks,which means they have higher credit risk.It is useless to use traditional approval process of retail loans for P2 P loan and we should find another way to manage the risk.It is worth noting that there have be a lot of historical trading data in P2 P lending market.A lot of big data analysis methods such as machine learning algorithm can be applied on P2 P default predicting.Credit scoring card is a success example of the combination of risk management and machine learning.“Fin-tech” is the trend of development.Base on the background above,this paper expects to provide some advice for domestic P2 P industry by analyzing foreign P2 P industry by empirical researching P2 P trading data.Research data is got from Lending club,which is the biggest P2 P lending market in U.S.The data has 153 dimensions of variables.The research object of this paper is to predict the default of P2 P trading in one year based on the application data.The period of research is from 2016Q1 to 2018Q2,which contains 1116659 records of P2 P trading.Firstly,the definition of default in one year is defined and the variables after the issue of loans are filtered.Then,this paper combines machine learning and P2 P risk management by apply XGBoost in predicting default,which is a frontier machine learning algorithm.Next,this paper uses professional methods to adjust the parameters of model and apply specialized test process to validate model.The test outcome shows that it is useful to predicting P2 P loan default in one year by using the model issued in this paper.In addition,this paper sort factors of default based on the model and find district of applicant is also an important factor.Finally,this paper provides some advice for domestic P2 P industry based on empirical research.Firstly,P2 P platforms should be encouraged to share P2 P trading data,which is a way to improve the level of risk management.Secondly,more information such as historical credit behaviors should be required to provide at the time of application to avoid asymmetric information.Then,an authoritative credit scoring frame should be developed in our country,which is the basis of risk management.Finally,different methods in predicting default should be combined in the whole process of P2 P loan to reduce the default rate.